Neural Network Based Throttle Actuator Model for Controller 2019-26-0247
HiL (Hardware-in-Loop) is a closed loop validation setup where some parts of the complete closed loop system are real hardware while the remaining parts of the loop are modelled. In our existing HiL setup, the controllers to be tested are real, while the remaining vehicle is modelled as plant model except for some actuators like throttle valve, waste-gate valve, injectors, etc. These actuators are connected to the HiL plant model via load-boxes. These actuators are not modelled as they are highly non-linear in nature and difficult to control. Hence, in this work, a neural network based real-time throttle actuator model for controller has been proposed. A throttle actuator contains a DC motor and spring-flap. But the main difficulty in modeling a throttle actuator is due to its nonlinear behavior. But, for HiL applications, a very accurate and real-time model is needed. For creating such an accurate ODE (ordinary differential equation) based model, parameter identification of each component of throttle needs to be done separately in general. But, for this individual parameter estimation of each component by dismantling the actuator, more effort and time is needed. Hence, a robust non-linear learning based model is designed using recurrent neural network. A new parameter estimation algorithm is also proposed by modifying meta-heuristic simulated annealing search algorithm. The proposed model is trained and validated offline by taking measurements from existing throttle actuator, including WLTP cycle. The test measurements are taken including all possible conditions of closing and opening of throttle actuator. Next the proposed model is evaluated offline in both closed and open loop. In both the cases, the proposed throttle actuator model has been found to be working satisfactorily, and, hence, the model is planned to be tested real-time on HiL as next step.